The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B5. Beijing 2008 
As we have noted above, for a prismatic building model the 
situation is rather simple, since the coarse geometry is described 
facades are typically planar polygons. We split the point cloud 
into groups with respect to the facades of the building using a 
simple buffer operation for each facade polygon. Each subset of 
the point cloud that is assigned to a particular facade is then 
interpolated into a regular raster. We can either us nearest 
neighborhood computation for the interpolation or triangulation. 
In general the methods used are similar to digital elevation 
models derived from aerial LIDAR. 
We refer to such a re-interpolated point cloud as a LASERMAP 
(Bohm, 2005). The term is composed from two terms describing 
the source of the data, a laser scanner, and the use of the data as 
a source for 2D mapping. Figure 3 shows a LASERMAP of the 
front facade of the aforementioned building. The gray values 
correspond to offsets relative to the plane of the facade. The 
map was computed at a resolution of 10 mm to preserve details, 
which gives an image of 2878 x 1778 pixels. Each pixel stores 
the offset in 16 bits. 
3.1 Generation of Normal Map and Displacement Map 
A LASERMAP is simply a grey value image and can be 
processed as such. An example for such a simple processing 
step is the generation of a normal map. The normal map stores 
the perturbations of the normal vector at each pixel to model the 
variation of the surface. The unit normal vectors for each pixel 
can be computed from the partial derivatives of the surface 
functions as represented by the LASERMAP. This is easily 
done applying derivative filters to the LASERMAP. Figure 
3 Figure 3 shows the rendering of a single facade polygon using 
a normal map derived from the above LASERMAP. While a 
normal map gives the impression of fine surface detail, this is 
only achieved by varying the shading of each output pixel; the 
actual geometry is still a flat polygon. This is advantageous as it 
does not increase the polygon count of the fully detailed model, 
when compared to the original model. However, the ‘flatness’ 
Figure 3. A LASERMAP of a single facade derived from 
the point cloud and its rendering as a normal map. 
of the surface is revealed to the observer under very oblique 
viewing angles. 
To overcome this ‘flatness’ true three-dimensional geometry 
has to be generated. This can only be achieved by adding 
vertices to the model and thereby generating new polygons. 
This is easily done by subdividing the façade polygon, a 
standard procedure in modelling. This procedure iteratively 
subdivides a polygon into smaller polygons, until the desired 
resolution is achieved. This procedure was initially suggested 
by (Catmull, 1974). The advantage of subdivision surfaces is 
that instead of generating vertices explicitly, they are generated 
implicitly, by storing the subdivision scheme and level. After 
the subdivision is defined the offset values stored in the 
LASERMAP are used to displace the generated vertices. The 
method is therefore referred to as displacement mapping. Figure 
4 shows the model mentioned before using displacement 
mapping. 
Figure 4. Rendering of a building model with 
displacement maps. 
4. SUBSTITUTION OF DEFECT AREAS 
Within most terrestrial laser scanning projects we frequently 
encounter scanning artifacts, which impair the quality of the 
point cloud. Such artifacts are often created by occlusions, as 
mentioned above, but can also be caused by varying surface 
reflectivity, beam deflections and other problems. It is rather 
difficult to correct these defects directly in the point cloud. 
However, it can be rather simple to treat these situations in a 
LASERMAP. 
Since facade architecture does not consist of purely random 
geometry, but is composed of repetitive elements, we can 
simply replace defective areas with a copy of an intact element. 
Since the representation of the LASERMAP is essentially the 
same as an image, image processing operations can be 
employed to automate this task. In figure 5 we show an example 
of a semi-automated repair process. The example is a detail 
from the faced already shown in figure 3. The right one of the 
two windows clearly has a defect, due to the window being half- 
opened at the time of scanning. The repair process starts by 
interactively marking the defective area in the LASERMAP, 
shown as a white box in the image. Then we automatically 
search for a similar area in the LASERMAP. This is 
implemented using simple template matching. We perform a 
global template matching across the full LASERMAP and the 
best match (depicted by a black box) is copied over the defect 
area. The result of the repair operation is shown in the bottom 
rows of figure 5. 
655